The MecCog Challenge: Agents collaboration to map Alzheimer’s disease mechanisms

Some research fields are hard to map, as their literature spans a wide range of studies built on differing—and sometimes conflicting—hypotheses. Alzheimer’s disease research is one such field. While the APOE4 allele is accepted as a major genetic risk factor for late-onset Alzheimer’s disease, its inundating ∼10,000-paper literature proposes at least forty distinct mechanisms. These can most likely be narrowed down to one or two but no reader can easily discern signal from noise on this scale. We introduce Mechanismus Cognito (MecCog), an open, community- driven initiative that turns such literature into a versioned, provenance-tracked dataset readable by both humans and agents. Instantiated here as a collaborative agent challenge, its objective is to create a curated corpus that is accessible to everyone. Agents claim a hypothesis, post to a shared board to coordinate with other agents and humans, and collate supporting papers and findings, always grounded in the published literature and enforced by verbatim quotes. They then use Pull Requests to promote vetted papers and quotes to the final paper list, trim irrelevant findings, and cross-review each other’s work. The end-goal is a versioned, consensus dataset that is both human and agent readable and can be subsequently evaluated to support or refute the hypotheses based on the existing literature. While instantiated here on APOE4/Alzheimer’s, the same collaborative challenge can be applied to any mechanism-mapping problem. As of 26 August 2026, 38 agents had posted 593 submissions and 4,038 board messages, selecting 490 paper–hypothesis pairs as evidence of the five hypotheses and 4,423 quotes.